Joe Witt created NIFI-7222:
------------------------------
Summary: FetchSFTP appears to not advise the remote system it is
done with a given resource resulting in too many open files
Key: NIFI-7222
URL: https://issues.apache.org/jira/browse/NIFI-7222
Project: Apache NiFi
Issue Type: Bug
Components: Extensions
Reporter: Joe Witt
Assignee: Joe Witt
Hi guys,
We have an issue with the FetchSFTP processor and the max open file
descriptors. In short, it seems that the FetchSFTP keeps the file open
“forever” on our Synology NAS, so we are reaching always the default max open
files limit of 1024 from our Synlogy NAS if we try to fetch 500’000 small 1MB
files (so in fact it’s not possible to read the files as everything is blocked
after 1024 files).
We found no option to rise the limit of max open files on the Synology NAS (but
that’s not NiFi’s fault 😉). We have also other linux machine with CentOS, but
the behavior there isn’t exactly always the same. Sometimes the file
descriptors get closed but sometimes as well not.
Synology has no lsof command, but this is how I’ve checked it:
user@nas-01:~$ sudo ls -l /proc/<SSHD SFTP process PID>/fd | wc -l
1024
Any comments how we can troubleshoot the issue?
Cheers Josef
Oh sorry, missed one of of the most important parts, we are using a 8-node
cluster with nifi 1.11.3 – so perfectly up to date.
Cheers Josef
Hi Joe
Ok, to our setup, we just bought a new powerful Synology NAS to use it as SFTP
server mainly for NiFi to replace our current SFTP linux machine. So the NAS is
empty and just configured for this single use case (read/write SFTP from NiFi).
Nothing else is running there at the moment. Important limit is per SSH/user
session ulimit -a 1024 open files max.:
root@nas-01:~# ulimit -a
core file size (blocks, -c) unlimited
data seg size (kbytes, -d) unlimited
scheduling priority (-e) 0
file size (blocks, -f) unlimited
pending signals (-i) 62025
max locked memory (kbytes, -l) 64
max memory size (kbytes, -m) unlimited
open files (-n) 1024
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) 819200
real-time priority (-r) 0
stack size (kbytes, -s) 8192
cpu time (seconds, -t) unlimited
max user processes (-u) 62025
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited
On NiFi side we are using an 8 node cluster, but it doesn’t matter whether I’m
using the whole cluster or just one single (primary) node. It’s clearly visible
that it’s related to the number of FetchSFTP processors running. So if I’m
distributing the load to 8 nodes I’m seeing 8 SFTP sessions on the NAS and we
can fetch 8x1024 files. I’m also seeing the file descriptors from each file
(per FetchSFTP processor = PID) on the NAS which has been fetched by NiFi. In
my understanding this files should be fetched and the file descriptor should be
closed after the transfer, but this doesn’t seems to be the case in most of the
times.
As soon as I’m stopping the “FetchSFTP” processor, the SFTP session seems to be
closed and all FDs are gone. So after stop/start I can fetch again 1024 files.
So I tried to troubleshoot a bit further and here is what I’ve done in NiFi and
on the NAS:
A screenshot of text
Description automatically generated
So I’ve done a ListSFTP and got 2880 flowfiles, they will be loadbalanced to
one single node (to simplify to test and only get 1 SFTP session on the NAS).
In the ControlRate I’m transferring every 10 seconds 10 flowfiles to the
FetchSFTP, that corelates directly with the open file descriptors on my NAS, as
you can see below. Sometimes, and I don’t know when or why, the SFTP session
will be closed and everything starts from scratch (not happened here) without
any notice on NiFi side. As you see, the FDs are growing with +10 every 10sec
and if I’m checking the path/filename of the open FDs I see that this are the
one which I’ve fetched.
root@nas-01:~# ps aux | grep sftp
root 1740 0.5 0.0 240848 8584 ? Ss 15:01 0:00 sshd:
ldr@internal-sftp
root 1753 0.0 0.0 23144 2360 pts/2 S+ 15:01 0:00 grep
--color=auto sftp
root 15520 0.0 0.0 241088 9252 ? Ss 13:38 0:02 sshd:
ldr@internal-sftp
root@nas-01:~#
root@nas-01:~# ls -l /proc/1740/fd | wc -l
24
root@nas-01:~# ls -l /proc/1740/fd | wc -l
34
root@nas-01:~# ls -l /proc/1740/fd | wc -l
44
root@nas-01:~# ls -l /proc/1740/fd | wc -l
54
root@nas-01:~# ls -l /proc/1740/fd | wc -l
64
root@p-li-nas-01:~# ls -l /proc/1740/fd | head
total 0
lr-x------ 1 root root 64 Mar 4 15:01 0 -> pipe:[1086218]
l-wx------ 1 root root 64 Mar 4 15:01 1 -> pipe:[1086219]
lr-x------+ 1 root root 64 Mar 4 15:01 10 ->
/volume1/test/2019-08-31/detail-20190831-0104-92.log.gz
lr-x------+ 1 root root 64 Mar 4 15:03 100 ->
/volume1/test/2019-08-31/detail-20190831-0052-91.log.gz
lr-x------+ 1 root root 64 Mar 4 15:03 101 ->
/volume1/test/2019-08-31/detail-20190831-0340-92.log.gz
lr-x------+ 1 root root 64 Mar 4 15:03 102 ->
/volume1/test/2019-08-31/detail-20190831-0246-91.log.gz
lr-x------+ 1 root root 64 Mar 4 15:03 103 ->
/volume1/test/2019-08-31/detail-20190831-0104-91.log.gz
lr-x------+ 1 root root 64 Mar 4 15:03 104 ->
/volume1/test/2019-08-31/detail-20190831-0150-91.log.gz
lr-x------+ 1 root root 64 Mar 4 15:03 105 ->
/volume1/test/2019-08-31/detail-20190831-0013-91.log.gz
…
So to sum up, one FetchSFTP generates one SFTP Session on the NAS. The SFTP
Session holds FDs which ,most of the time, doesn’t get closed. Reproduceable
with the template above with FetchSFTP to a CentOS machine or a Synology NAS.
Main question is now, why were the FDs not closed or when should the SFTP
session gets closed.
Thanks
Just checked the open sockets on the NiFi machine where FetchSFTP is running,
of course there is just one SSH session if I’m using just one single processor…
So the SFTP transfer is hidden in the SSH session.
Open TCP sessions on NiFi
[user@ nifi-05 ~]$ netstat -vatn | grep x.y.z.232
tcp 0 0 x.y.z.144:33628 x.y.z.232:22 ESTABLISHED
Any comments are welcome. Still unclear where the open FDs on the NAS (SFTP
server) are coming from or how it should work from NiFi perspective.
Cheers Josef
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
